Research On Dynamic Multi-Objective Evolutionary Algorithm Based On Prediction And Its Application

Dynamic multi-objective optimization problems are widely existed in real life.When using evolutionary algorithm to solve dynamic multi-objective optimization problems,we should not only consider the conflict and coupling between research objectives,as well as the changes of objective function and decision variables over time,but also require the algorithm to track Pareto front in the new environment quickly and accurately.The convergence and diversity of prediction individuals generated by dynamic multi-objective optimization algorithm based on prediction need to be improved,and the algorithm does not consider the change of shape of Pareto frontier with time when using historical information to predict new individuals.Therefore,the prediction method is studied in this paper.Aiming at the problems of poor individual performance prediction and dynamic shape change of Pareto frontier,two dynamic multi-objective optimization algorithms are proposed and applied to resource scheduling in virtual dynamic cloud environment.The research content of this paper is mainly divided into the following three aspects:(1)To solve the problem of poor individual performance,a dynam ic multi-objective evolutionary algorithm based on weight vector clustering is proposed.Firstly,a uniform weight vector is generated in the target space,the individuals in the population are clustered,their clustering centers are calculated,and a time series is established for the clustering centers.Secondly,the same weight vector is used to supplement the individual with corresponding coping strategies according to different clustering situations,and the difference model is used to predict the indi vidual.Finally,the individual supplement strategy is introduced to make full use of historical information.In order to verify the performance of the proposed algorithm,it is simulated and compared with four representative algorithms.The experimental r esults show that the improved strategy can improve the ability of the algorithm to respond to environmental changes,and it can be seen that the algorithm can improve the performance of the prediction individual and solve the dynamic multi-objective optimization problem well.(2)Aiming at the dynamic change of Pareto frontier shape with time in different environments,a dynamic multi-objective evolutionary algorithm based on Pareto frontier shape prediction is proposed in this paper.Firstly,the algorithm judges the shape of Pareto frontier in the new environment when the environment changes,divides the shape of Pareto frontier into concave curve,convex curve and straight line,and adopts different weight vector generation methods according to the judged shape of Pareto frontier.Secondly,the individuals in the population are clustered based on the weight vector,and the coping strategies are selected adaptively.Finally,the proposed algorithm is compared with the classical algorithm on FDA1-FDA5,DMOP and Fun series test functions.Experimental results show that the algorithm can solve the problem of Pareto frontier shape changing with time well.(3)In the cloud environment resource scheduling problem,how to make the cloud environment resource scheduling in speed,efficiency and scheduling scheme to meet the needs of the rapid development of the society is very important.The research focus of the cloud environment resource scheduling problem is to solve the mapping between virtual machines and physical hosts while solving the load balancing problem of the system.Due to the dynamic change of CPU resources required by virtual machine with time,it has problems of poor individual quality prediction and Pareto frontier change with time.Therefore,a dynamic cloud resource schedul ing algorithm based on virtual machine migration is proposed by combining the above two algorithms.In the coding stage,the algorithm adopts the real coding mode.In the static algorithm,crossover and mutation respectively use the midpoint crossover mode and gene probability density variation mode.In the environmental response stage,individuals were clustered and predicted by means of gene niche mode selection and individual translation.By setting five models and comparing the proposed algorithm with the four comparison algorithms,it is proved that the proposed algorithm can solve the resource scheduling problem in the cloud environment well and the performance of the algorithm is better than the other four comparison algorithms.